I've made some good progress with wasm. I've compiled the basic parts of flume.c code I've been writing to wasm, and I just tested it with a node script around it. Node is handling the IO. node reads 64k blocks into the wasm space, then I use wasm to scan for the values. writing output is somewhat slower, but I can scan the entire database for one thread in half a second. This is very good news! It means it's possible to incrementally deploy bits of things written in wasm, and boost perf, without having to deal with another native addon.